fs/nfs/nfs4state.c

   1 /*
   2  *  fs/nfs/nfs4state.c
   3  *
   4  *  Client-side XDR for NFSv4.
   5  *
   6  *  Copyright (c) 2002 The Regents of the University of Michigan.
   7  *  All rights reserved.
   8  *
   9  *  Kendrick Smith <kmsmith@umich.edu>
  10  *
  11  *  Redistribution and use in source and binary forms, with or without
  12  *  modification, are permitted provided that the following conditions
  13  *  are met:
  14  *
  15  *  1. Redistributions of source code must retain the above copyright
  16  *     notice, this list of conditions and the following disclaimer.
  17  *  2. Redistributions in binary form must reproduce the above copyright
  18  *     notice, this list of conditions and the following disclaimer in the
  19  *     documentation and/or other materials provided with the distribution.
  20  *  3. Neither the name of the University nor the names of its
  21  *     contributors may be used to endorse or promote products derived
  22  *     from this software without specific prior written permission.
  23  *
  24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  35  *
  36  * Implementation of the NFSv4 state model.  For the time being,
  37  * this is minimal, but will be made much more complex in a
  38  * subsequent patch.
  39  */
  40
  41 #include <linux/kernel.h>
  42 #include <linux/slab.h>
  43 #include <linux/fs.h>
  44 #include <linux/nfs_fs.h>
  45 #include <linux/kthread.h>
  46 #include <linux/module.h>
  47 #include <linux/random.h>
  48 #include <linux/ratelimit.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/bitops.h>
  51 #include <linux/jiffies.h>
  52
  53 #include <linux/sunrpc/clnt.h>
  54
  55 #include "nfs4_fs.h"
  56 #include "callback.h"
  57 #include "delegation.h"
  58 #include "internal.h"
  59 #include "nfs4idmap.h"
  60 #include "nfs4session.h"
  61 #include "pnfs.h"
  62 #include "netns.h"
  63
  64 #define NFSDBG_FACILITY         NFSDBG_STATE
  65
  66 #define OPENOWNER_POOL_SIZE     8
  67
  68 const nfs4_stateid zero_stateid = {
  69         { .data = { 0 } },
  70         .type = NFS4_SPECIAL_STATEID_TYPE,
  71 };
  72 static DEFINE_MUTEX(nfs_clid_init_mutex);
  73
  74 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
  75 {
  76         struct nfs4_setclientid_res clid = {
  77                 .clientid = clp->cl_clientid,
  78                 .confirm = clp->cl_confirm,
  79         };
  80         unsigned short port;
  81         int status;
  82         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
  83
  84         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
  85                 goto do_confirm;
  86         port = nn->nfs_callback_tcpport;
  87         if (clp->cl_addr.ss_family == AF_INET6)
  88                 port = nn->nfs_callback_tcpport6;
  89
  90         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
  91         if (status != 0)
  92                 goto out;
  93         clp->cl_clientid = clid.clientid;
  94         clp->cl_confirm = clid.confirm;
  95         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
  96 do_confirm:
  97         status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
  98         if (status != 0)
  99                 goto out;
 100         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 101         nfs4_schedule_state_renewal(clp);
 102 out:
 103         return status;
 104 }
 105
 106 /**
 107  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
 108  *
 109  * @clp: nfs_client under test
 110  * @result: OUT: found nfs_client, or clp
 111  * @cred: credential to use for trunking test
 112  *
 113  * Returns zero, a negative errno, or a negative NFS4ERR status.
 114  * If zero is returned, an nfs_client pointer is planted in
 115  * "result".
 116  *
 117  * Note: The returned client may not yet be marked ready.
 118  */
 119 int nfs40_discover_server_trunking(struct nfs_client *clp,
 120                                    struct nfs_client **result,
 121                                    struct rpc_cred *cred)
 122 {
 123         struct nfs4_setclientid_res clid = {
 124                 .clientid = clp->cl_clientid,
 125                 .confirm = clp->cl_confirm,
 126         };
 127         struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
 128         unsigned short port;
 129         int status;
 130
 131         port = nn->nfs_callback_tcpport;
 132         if (clp->cl_addr.ss_family == AF_INET6)
 133                 port = nn->nfs_callback_tcpport6;
 134
 135         status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
 136         if (status != 0)
 137                 goto out;
 138         clp->cl_clientid = clid.clientid;
 139         clp->cl_confirm = clid.confirm;
 140
 141         status = nfs40_walk_client_list(clp, result, cred);
 142         if (status == 0) {
 143                 /* Sustain the lease, even if it's empty.  If the clientid4
 144                  * goes stale it's of no use for trunking discovery. */
 145                 nfs4_schedule_state_renewal(*result);
 146         }
 147 out:
 148         return status;
 149 }
 150
 151 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
 152 {
 153         struct rpc_cred *cred = NULL;
 154
 155         if (clp->cl_machine_cred != NULL)
 156                 cred = get_rpccred(clp->cl_machine_cred);
 157         return cred;
 158 }
 159
 160 static void nfs4_root_machine_cred(struct nfs_client *clp)
 161 {
 162         struct rpc_cred *cred, *new;
 163
 164         new = rpc_lookup_machine_cred(NULL);
 165         spin_lock(&clp->cl_lock);
 166         cred = clp->cl_machine_cred;
 167         clp->cl_machine_cred = new;
 168         spin_unlock(&clp->cl_lock);
 169         if (cred != NULL)
 170                 put_rpccred(cred);
 171 }
 172
 173 static struct rpc_cred *
 174 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
 175 {
 176         struct rpc_cred *cred = NULL;
 177         struct nfs4_state_owner *sp;
 178         struct rb_node *pos;
 179
 180         for (pos = rb_first(&server->state_owners);
 181              pos != NULL;
 182              pos = rb_next(pos)) {
 183                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
 184                 if (list_empty(&sp->so_states))
 185                         continue;
 186                 cred = get_rpccred(sp->so_cred);
 187                 break;
 188         }
 189         return cred;
 190 }
 191
 192 /**
 193  * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
 194  * @clp: client state handle
 195  *
 196  * Returns an rpc_cred with reference count bumped, or NULL.
 197  * Caller must hold clp->cl_lock.
 198  */
 199 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
 200 {
 201         struct rpc_cred *cred = NULL;
 202         struct nfs_server *server;
 203
 204         /* Use machine credentials if available */
 205         cred = nfs4_get_machine_cred_locked(clp);
 206         if (cred != NULL)
 207                 goto out;
 208
 209         rcu_read_lock();
 210         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 211                 cred = nfs4_get_renew_cred_server_locked(server);
 212                 if (cred != NULL)
 213                         break;
 214         }
 215         rcu_read_unlock();
 216
 217 out:
 218         return cred;
 219 }
 220
 221 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
 222 {
 223         if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
 224                 spin_lock(&tbl->slot_tbl_lock);
 225                 nfs41_wake_slot_table(tbl);
 226                 spin_unlock(&tbl->slot_tbl_lock);
 227         }
 228 }
 229
 230 static void nfs4_end_drain_session(struct nfs_client *clp)
 231 {
 232         struct nfs4_session *ses = clp->cl_session;
 233
 234         if (clp->cl_slot_tbl) {
 235                 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
 236                 return;
 237         }
 238
 239         if (ses != NULL) {
 240                 nfs4_end_drain_slot_table(&ses->bc_slot_table);
 241                 nfs4_end_drain_slot_table(&ses->fc_slot_table);
 242         }
 243 }
 244
 245 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
 246 {
 247         set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
 248         spin_lock(&tbl->slot_tbl_lock);
 249         if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
 250                 reinit_completion(&tbl->complete);
 251                 spin_unlock(&tbl->slot_tbl_lock);
 252                 return wait_for_completion_interruptible(&tbl->complete);
 253         }
 254         spin_unlock(&tbl->slot_tbl_lock);
 255         return 0;
 256 }
 257
 258 static int nfs4_begin_drain_session(struct nfs_client *clp)
 259 {
 260         struct nfs4_session *ses = clp->cl_session;
 261         int ret = 0;
 262
 263         if (clp->cl_slot_tbl)
 264                 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
 265
 266         /* back channel */
 267         ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
 268         if (ret)
 269                 return ret;
 270         /* fore channel */
 271         return nfs4_drain_slot_tbl(&ses->fc_slot_table);
 272 }
 273
 274 #if defined(CONFIG_NFS_V4_1)
 275
 276 static int nfs41_setup_state_renewal(struct nfs_client *clp)
 277 {
 278         int status;
 279         struct nfs_fsinfo fsinfo;
 280         unsigned long now;
 281
 282         if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
 283                 nfs4_schedule_state_renewal(clp);
 284                 return 0;
 285         }
 286
 287         now = jiffies;
 288         status = nfs4_proc_get_lease_time(clp, &fsinfo);
 289         if (status == 0) {
 290                 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
 291                 nfs4_schedule_state_renewal(clp);
 292         }
 293
 294         return status;
 295 }
 296
 297 static void nfs41_finish_session_reset(struct nfs_client *clp)
 298 {
 299         clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 300         clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 301         /* create_session negotiated new slot table */
 302         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 303         nfs41_setup_state_renewal(clp);
 304 }
 305
 306 int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
 307 {
 308         int status;
 309
 310         if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
 311                 goto do_confirm;
 312         status = nfs4_proc_exchange_id(clp, cred);
 313         if (status != 0)
 314                 goto out;
 315         set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 316 do_confirm:
 317         status = nfs4_proc_create_session(clp, cred);
 318         if (status != 0)
 319                 goto out;
 320         nfs41_finish_session_reset(clp);
 321         nfs_mark_client_ready(clp, NFS_CS_READY);
 322 out:
 323         return status;
 324 }
 325
 326 /**
 327  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
 328  *
 329  * @clp: nfs_client under test
 330  * @result: OUT: found nfs_client, or clp
 331  * @cred: credential to use for trunking test
 332  *
 333  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
 334  * If NFS4_OK is returned, an nfs_client pointer is planted in
 335  * "result".
 336  *
 337  * Note: The returned client may not yet be marked ready.
 338  */
 339 int nfs41_discover_server_trunking(struct nfs_client *clp,
 340                                    struct nfs_client **result,
 341                                    struct rpc_cred *cred)
 342 {
 343         int status;
 344
 345         status = nfs4_proc_exchange_id(clp, cred);
 346         if (status != NFS4_OK)
 347                 return status;
 348
 349         status = nfs41_walk_client_list(clp, result, cred);
 350         if (status < 0)
 351                 return status;
 352         if (clp != *result)
 353                 return 0;
 354
 355         /* Purge state if the client id was established in a prior instance */
 356         if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
 357                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
 358         else
 359                 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 360         nfs4_schedule_state_manager(clp);
 361         status = nfs_wait_client_init_complete(clp);
 362         if (status < 0)
 363                 nfs_put_client(clp);
 364         return status;
 365 }
 366
 367 #endif /* CONFIG_NFS_V4_1 */
 368
 369 /**
 370  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
 371  * @clp: client state handle
 372  *
 373  * Returns an rpc_cred with reference count bumped, or NULL.
 374  */
 375 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
 376 {
 377         struct rpc_cred *cred;
 378
 379         spin_lock(&clp->cl_lock);
 380         cred = nfs4_get_machine_cred_locked(clp);
 381         spin_unlock(&clp->cl_lock);
 382         return cred;
 383 }
 384
 385 static struct nfs4_state_owner *
 386 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
 387 {
 388         struct rb_node **p = &server->state_owners.rb_node,
 389                        *parent = NULL;
 390         struct nfs4_state_owner *sp;
 391
 392         while (*p != NULL) {
 393                 parent = *p;
 394                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 395
 396                 if (cred < sp->so_cred)
 397                         p = &parent->rb_left;
 398                 else if (cred > sp->so_cred)
 399                         p = &parent->rb_right;
 400                 else {
 401                         if (!list_empty(&sp->so_lru))
 402                                 list_del_init(&sp->so_lru);
 403                         atomic_inc(&sp->so_count);
 404                         return sp;
 405                 }
 406         }
 407         return NULL;
 408 }
 409
 410 static struct nfs4_state_owner *
 411 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
 412 {
 413         struct nfs_server *server = new->so_server;
 414         struct rb_node **p = &server->state_owners.rb_node,
 415                        *parent = NULL;
 416         struct nfs4_state_owner *sp;
 417         int err;
 418
 419         while (*p != NULL) {
 420                 parent = *p;
 421                 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
 422
 423                 if (new->so_cred < sp->so_cred)
 424                         p = &parent->rb_left;
 425                 else if (new->so_cred > sp->so_cred)
 426                         p = &parent->rb_right;
 427                 else {
 428                         if (!list_empty(&sp->so_lru))
 429                                 list_del_init(&sp->so_lru);
 430                         atomic_inc(&sp->so_count);
 431                         return sp;
 432                 }
 433         }
 434         err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
 435         if (err)
 436                 return ERR_PTR(err);
 437         rb_link_node(&new->so_server_node, parent, p);
 438         rb_insert_color(&new->so_server_node, &server->state_owners);
 439         return new;
 440 }
 441
 442 static void
 443 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
 444 {
 445         struct nfs_server *server = sp->so_server;
 446
 447         if (!RB_EMPTY_NODE(&sp->so_server_node))
 448                 rb_erase(&sp->so_server_node, &server->state_owners);
 449         ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
 450 }
 451
 452 static void
 453 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
 454 {
 455         sc->create_time = ktime_get();
 456         sc->flags = 0;
 457         sc->counter = 0;
 458         spin_lock_init(&sc->lock);
 459         INIT_LIST_HEAD(&sc->list);
 460         rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
 461 }
 462
 463 static void
 464 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
 465 {
 466         rpc_destroy_wait_queue(&sc->wait);
 467 }
 468
 469 /*
 470  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
 471  * create a new state_owner.
 472  *
 473  */
 474 static struct nfs4_state_owner *
 475 nfs4_alloc_state_owner(struct nfs_server *server,
 476                 struct rpc_cred *cred,
 477                 gfp_t gfp_flags)
 478 {
 479         struct nfs4_state_owner *sp;
 480
 481         sp = kzalloc(sizeof(*sp), gfp_flags);
 482         if (!sp)
 483                 return NULL;
 484         sp->so_server = server;
 485         sp->so_cred = get_rpccred(cred);
 486         spin_lock_init(&sp->so_lock);
 487         INIT_LIST_HEAD(&sp->so_states);
 488         nfs4_init_seqid_counter(&sp->so_seqid);
 489         atomic_set(&sp->so_count, 1);
 490         INIT_LIST_HEAD(&sp->so_lru);
 491         seqcount_init(&sp->so_reclaim_seqcount);
 492         mutex_init(&sp->so_delegreturn_mutex);
 493         return sp;
 494 }
 495
 496 static void
 497 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
 498 {
 499         struct rb_node *rb_node = &sp->so_server_node;
 500
 501         if (!RB_EMPTY_NODE(rb_node)) {
 502                 struct nfs_server *server = sp->so_server;
 503                 struct nfs_client *clp = server->nfs_client;
 504
 505                 spin_lock(&clp->cl_lock);
 506                 if (!RB_EMPTY_NODE(rb_node)) {
 507                         rb_erase(rb_node, &server->state_owners);
 508                         RB_CLEAR_NODE(rb_node);
 509                 }
 510                 spin_unlock(&clp->cl_lock);
 511         }
 512 }
 513
 514 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
 515 {
 516         nfs4_destroy_seqid_counter(&sp->so_seqid);
 517         put_rpccred(sp->so_cred);
 518         kfree(sp);
 519 }
 520
 521 static void nfs4_gc_state_owners(struct nfs_server *server)
 522 {
 523         struct nfs_client *clp = server->nfs_client;
 524         struct nfs4_state_owner *sp, *tmp;
 525         unsigned long time_min, time_max;
 526         LIST_HEAD(doomed);
 527
 528         spin_lock(&clp->cl_lock);
 529         time_max = jiffies;
 530         time_min = (long)time_max - (long)clp->cl_lease_time;
 531         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 532                 /* NB: LRU is sorted so that oldest is at the head */
 533                 if (time_in_range(sp->so_expires, time_min, time_max))
 534                         break;
 535                 list_move(&sp->so_lru, &doomed);
 536                 nfs4_remove_state_owner_locked(sp);
 537         }
 538         spin_unlock(&clp->cl_lock);
 539
 540         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 541                 list_del(&sp->so_lru);
 542                 nfs4_free_state_owner(sp);
 543         }
 544 }
 545
 546 /**
 547  * nfs4_get_state_owner - Look up a state owner given a credential
 548  * @server: nfs_server to search
 549  * @cred: RPC credential to match
 550  *
 551  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
 552  */
 553 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
 554                                               struct rpc_cred *cred,
 555                                               gfp_t gfp_flags)
 556 {
 557         struct nfs_client *clp = server->nfs_client;
 558         struct nfs4_state_owner *sp, *new;
 559
 560         spin_lock(&clp->cl_lock);
 561         sp = nfs4_find_state_owner_locked(server, cred);
 562         spin_unlock(&clp->cl_lock);
 563         if (sp != NULL)
 564                 goto out;
 565         new = nfs4_alloc_state_owner(server, cred, gfp_flags);
 566         if (new == NULL)
 567                 goto out;
 568         do {
 569                 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
 570                         break;
 571                 spin_lock(&clp->cl_lock);
 572                 sp = nfs4_insert_state_owner_locked(new);
 573                 spin_unlock(&clp->cl_lock);
 574         } while (sp == ERR_PTR(-EAGAIN));
 575         if (sp != new)
 576                 nfs4_free_state_owner(new);
 577 out:
 578         nfs4_gc_state_owners(server);
 579         return sp;
 580 }
 581
 582 /**
 583  * nfs4_put_state_owner - Release a nfs4_state_owner
 584  * @sp: state owner data to release
 585  *
 586  * Note that we keep released state owners on an LRU
 587  * list.
 588  * This caches valid state owners so that they can be
 589  * reused, to avoid the OPEN_CONFIRM on minor version 0.
 590  * It also pins the uniquifier of dropped state owners for
 591  * a while, to ensure that those state owner names are
 592  * never reused.
 593  */
 594 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
 595 {
 596         struct nfs_server *server = sp->so_server;
 597         struct nfs_client *clp = server->nfs_client;
 598
 599         if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
 600                 return;
 601
 602         sp->so_expires = jiffies;
 603         list_add_tail(&sp->so_lru, &server->state_owners_lru);
 604         spin_unlock(&clp->cl_lock);
 605 }
 606
 607 /**
 608  * nfs4_purge_state_owners - Release all cached state owners
 609  * @server: nfs_server with cached state owners to release
 610  *
 611  * Called at umount time.  Remaining state owners will be on
 612  * the LRU with ref count of zero.
 613  */
 614 void nfs4_purge_state_owners(struct nfs_server *server)
 615 {
 616         struct nfs_client *clp = server->nfs_client;
 617         struct nfs4_state_owner *sp, *tmp;
 618         LIST_HEAD(doomed);
 619
 620         spin_lock(&clp->cl_lock);
 621         list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
 622                 list_move(&sp->so_lru, &doomed);
 623                 nfs4_remove_state_owner_locked(sp);
 624         }
 625         spin_unlock(&clp->cl_lock);
 626
 627         list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
 628                 list_del(&sp->so_lru);
 629                 nfs4_free_state_owner(sp);
 630         }
 631 }
 632
 633 static struct nfs4_state *
 634 nfs4_alloc_open_state(void)
 635 {
 636         struct nfs4_state *state;
 637
 638         state = kzalloc(sizeof(*state), GFP_NOFS);
 639         if (!state)
 640                 return NULL;
 641         atomic_set(&state->count, 1);
 642         INIT_LIST_HEAD(&state->lock_states);
 643         spin_lock_init(&state->state_lock);
 644         seqlock_init(&state->seqlock);
 645         return state;
 646 }
 647
 648 void
 649 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
 650 {
 651         if (state->state == fmode)
 652                 return;
 653         /* NB! List reordering - see the reclaim code for why.  */
 654         if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
 655                 if (fmode & FMODE_WRITE)
 656                         list_move(&state->open_states, &state->owner->so_states);
 657                 else
 658                         list_move_tail(&state->open_states, &state->owner->so_states);
 659         }
 660         state->state = fmode;
 661 }
 662
 663 static struct nfs4_state *
 664 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
 665 {
 666         struct nfs_inode *nfsi = NFS_I(inode);
 667         struct nfs4_state *state;
 668
 669         list_for_each_entry(state, &nfsi->open_states, inode_states) {
 670                 if (state->owner != owner)
 671                         continue;
 672                 if (!nfs4_valid_open_stateid(state))
 673                         continue;
 674                 if (atomic_inc_not_zero(&state->count))
 675                         return state;
 676         }
 677         return NULL;
 678 }
 679
 680 static void
 681 nfs4_free_open_state(struct nfs4_state *state)
 682 {
 683         kfree(state);
 684 }
 685
 686 struct nfs4_state *
 687 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
 688 {
 689         struct nfs4_state *state, *new;
 690         struct nfs_inode *nfsi = NFS_I(inode);
 691
 692         spin_lock(&inode->i_lock);
 693         state = __nfs4_find_state_byowner(inode, owner);
 694         spin_unlock(&inode->i_lock);
 695         if (state)
 696                 goto out;
 697         new = nfs4_alloc_open_state();
 698         spin_lock(&owner->so_lock);
 699         spin_lock(&inode->i_lock);
 700         state = __nfs4_find_state_byowner(inode, owner);
 701         if (state == NULL && new != NULL) {
 702                 state = new;
 703                 state->owner = owner;
 704                 atomic_inc(&owner->so_count);
 705                 list_add(&state->inode_states, &nfsi->open_states);
 706                 ihold(inode);
 707                 state->inode = inode;
 708                 spin_unlock(&inode->i_lock);
 709                 /* Note: The reclaim code dictates that we add stateless
 710                  * and read-only stateids to the end of the list */
 711                 list_add_tail(&state->open_states, &owner->so_states);
 712                 spin_unlock(&owner->so_lock);
 713         } else {
 714                 spin_unlock(&inode->i_lock);
 715                 spin_unlock(&owner->so_lock);
 716                 if (new)
 717                         nfs4_free_open_state(new);
 718         }
 719 out:
 720         return state;
 721 }
 722
 723 void nfs4_put_open_state(struct nfs4_state *state)
 724 {
 725         struct inode *inode = state->inode;
 726         struct nfs4_state_owner *owner = state->owner;
 727
 728         if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
 729                 return;
 730         spin_lock(&inode->i_lock);
 731         list_del(&state->inode_states);
 732         list_del(&state->open_states);
 733         spin_unlock(&inode->i_lock);
 734         spin_unlock(&owner->so_lock);
 735         iput(inode);
 736         nfs4_free_open_state(state);
 737         nfs4_put_state_owner(owner);
 738 }
 739
 740 /*
 741  * Close the current file.
 742  */
 743 static void __nfs4_close(struct nfs4_state *state,
 744                 fmode_t fmode, gfp_t gfp_mask, int wait)
 745 {
 746         struct nfs4_state_owner *owner = state->owner;
 747         int call_close = 0;
 748         fmode_t newstate;
 749
 750         atomic_inc(&owner->so_count);
 751         /* Protect against nfs4_find_state() */
 752         spin_lock(&owner->so_lock);
 753         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
 754                 case FMODE_READ:
 755                         state->n_rdonly--;
 756                         break;
 757                 case FMODE_WRITE:
 758                         state->n_wronly--;
 759                         break;
 760                 case FMODE_READ|FMODE_WRITE:
 761                         state->n_rdwr--;
 762         }
 763         newstate = FMODE_READ|FMODE_WRITE;
 764         if (state->n_rdwr == 0) {
 765                 if (state->n_rdonly == 0) {
 766                         newstate &= ~FMODE_READ;
 767                         call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
 768                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 769                 }
 770                 if (state->n_wronly == 0) {
 771                         newstate &= ~FMODE_WRITE;
 772                         call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
 773                         call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
 774                 }
 775                 if (newstate == 0)
 776                         clear_bit(NFS_DELEGATED_STATE, &state->flags);
 777         }
 778         nfs4_state_set_mode_locked(state, newstate);
 779         spin_unlock(&owner->so_lock);
 780
 781         if (!call_close) {
 782                 nfs4_put_open_state(state);
 783                 nfs4_put_state_owner(owner);
 784         } else
 785                 nfs4_do_close(state, gfp_mask, wait);
 786 }
 787
 788 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
 789 {
 790         __nfs4_close(state, fmode, GFP_NOFS, 0);
 791 }
 792
 793 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
 794 {
 795         __nfs4_close(state, fmode, GFP_KERNEL, 1);
 796 }
 797
 798 /*
 799  * Search the state->lock_states for an existing lock_owner
 800  * that is compatible with current->files
 801  */
 802 static struct nfs4_lock_state *
 803 __nfs4_find_lock_state(struct nfs4_state *state,
 804                        fl_owner_t fl_owner, fl_owner_t fl_owner2)
 805 {
 806         struct nfs4_lock_state *pos;
 807         list_for_each_entry(pos, &state->lock_states, ls_locks) {
 808                 if (pos->ls_owner != fl_owner &&
 809                     pos->ls_owner != fl_owner2)
 810                         continue;
 811                 atomic_inc(&pos->ls_count);
 812                 return pos;
 813         }
 814         return NULL;
 815 }
 816
 817 /*
 818  * Return a compatible lock_state. If no initialized lock_state structure
 819  * exists, return an uninitialized one.
 820  *
 821  */
 822 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
 823 {
 824         struct nfs4_lock_state *lsp;
 825         struct nfs_server *server = state->owner->so_server;
 826
 827         lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
 828         if (lsp == NULL)
 829                 return NULL;
 830         nfs4_init_seqid_counter(&lsp->ls_seqid);
 831         atomic_set(&lsp->ls_count, 1);
 832         lsp->ls_state = state;
 833         lsp->ls_owner = fl_owner;
 834         lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
 835         if (lsp->ls_seqid.owner_id < 0)
 836                 goto out_free;
 837         INIT_LIST_HEAD(&lsp->ls_locks);
 838         return lsp;
 839 out_free:
 840         kfree(lsp);
 841         return NULL;
 842 }
 843
 844 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
 845 {
 846         ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
 847         nfs4_destroy_seqid_counter(&lsp->ls_seqid);
 848         kfree(lsp);
 849 }
 850
 851 /*
 852  * Return a compatible lock_state. If no initialized lock_state structure
 853  * exists, return an uninitialized one.
 854  *
 855  */
 856 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
 857 {
 858         struct nfs4_lock_state *lsp, *new = NULL;
 859
 860         for(;;) {
 861                 spin_lock(&state->state_lock);
 862                 lsp = __nfs4_find_lock_state(state, owner, 0);
 863                 if (lsp != NULL)
 864                         break;
 865                 if (new != NULL) {
 866                         list_add(&new->ls_locks, &state->lock_states);
 867                         set_bit(LK_STATE_IN_USE, &state->flags);
 868                         lsp = new;
 869                         new = NULL;
 870                         break;
 871                 }
 872                 spin_unlock(&state->state_lock);
 873                 new = nfs4_alloc_lock_state(state, owner);
 874                 if (new == NULL)
 875                         return NULL;
 876         }
 877         spin_unlock(&state->state_lock);
 878         if (new != NULL)
 879                 nfs4_free_lock_state(state->owner->so_server, new);
 880         return lsp;
 881 }
 882
 883 /*
 884  * Release reference to lock_state, and free it if we see that
 885  * it is no longer in use
 886  */
 887 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
 888 {
 889         struct nfs_server *server;
 890         struct nfs4_state *state;
 891
 892         if (lsp == NULL)
 893                 return;
 894         state = lsp->ls_state;
 895         if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
 896                 return;
 897         list_del(&lsp->ls_locks);
 898         if (list_empty(&state->lock_states))
 899                 clear_bit(LK_STATE_IN_USE, &state->flags);
 900         spin_unlock(&state->state_lock);
 901         server = state->owner->so_server;
 902         if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
 903                 struct nfs_client *clp = server->nfs_client;
 904
 905                 clp->cl_mvops->free_lock_state(server, lsp);
 906         } else
 907                 nfs4_free_lock_state(server, lsp);
 908 }
 909
 910 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
 911 {
 912         struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
 913
 914         dst->fl_u.nfs4_fl.owner = lsp;
 915         atomic_inc(&lsp->ls_count);
 916 }
 917
 918 static void nfs4_fl_release_lock(struct file_lock *fl)
 919 {
 920         nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
 921 }
 922
 923 static const struct file_lock_operations nfs4_fl_lock_ops = {
 924         .fl_copy_lock = nfs4_fl_copy_lock,
 925         .fl_release_private = nfs4_fl_release_lock,
 926 };
 927
 928 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
 929 {
 930         struct nfs4_lock_state *lsp;
 931
 932         if (fl->fl_ops != NULL)
 933                 return 0;
 934         lsp = nfs4_get_lock_state(state, fl->fl_owner);
 935         if (lsp == NULL)
 936                 return -ENOMEM;
 937         fl->fl_u.nfs4_fl.owner = lsp;
 938         fl->fl_ops = &nfs4_fl_lock_ops;
 939         return 0;
 940 }
 941
 942 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
 943                 struct nfs4_state *state,
 944                 const struct nfs_lock_context *l_ctx)
 945 {
 946         struct nfs4_lock_state *lsp;
 947         fl_owner_t fl_owner, fl_flock_owner;
 948         int ret = -ENOENT;
 949
 950         if (l_ctx == NULL)
 951                 goto out;
 952
 953         if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
 954                 goto out;
 955
 956         fl_owner = l_ctx->lockowner;
 957         fl_flock_owner = l_ctx->open_context->flock_owner;
 958
 959         spin_lock(&state->state_lock);
 960         lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
 961         if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
 962                 ret = -EIO;
 963         else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
 964                 nfs4_stateid_copy(dst, &lsp->ls_stateid);
 965                 ret = 0;
 966         }
 967         spin_unlock(&state->state_lock);
 968         nfs4_put_lock_state(lsp);
 969 out:
 970         return ret;
 971 }
 972
 973 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
 974 {
 975         const nfs4_stateid *src;
 976         int seq;
 977
 978         do {
 979                 src = &zero_stateid;
 980                 seq = read_seqbegin(&state->seqlock);
 981                 if (test_bit(NFS_OPEN_STATE, &state->flags))
 982                         src = &state->open_stateid;
 983                 nfs4_stateid_copy(dst, src);
 984         } while (read_seqretry(&state->seqlock, seq));
 985 }
 986
 987 /*
 988  * Byte-range lock aware utility to initialize the stateid of read/write
 989  * requests.
 990  */
 991 int nfs4_select_rw_stateid(struct nfs4_state *state,
 992                 fmode_t fmode, const struct nfs_lock_context *l_ctx,
 993                 nfs4_stateid *dst, struct rpc_cred **cred)
 994 {
 995         int ret;
 996
 997         if (!nfs4_valid_open_stateid(state))
 998                 return -EIO;
 999         if (cred != NULL)
1000                 *cred = NULL;
1001         ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1002         if (ret == -EIO)
1003                 /* A lost lock - don't even consider delegations */
1004                 goto out;
1005         /* returns true if delegation stateid found and copied */
1006         if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1007                 ret = 0;
1008                 goto out;
1009         }
1010         if (ret != -ENOENT)
1011                 /* nfs4_copy_delegation_stateid() didn't over-write
1012                  * dst, so it still has the lock stateid which we now
1013                  * choose to use.
1014                  */
1015                 goto out;
1016         nfs4_copy_open_stateid(dst, state);
1017         ret = 0;
1018 out:
1019         if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1020                 dst->seqid = 0;
1021         return ret;
1022 }
1023
1024 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1025 {
1026         struct nfs_seqid *new;
1027
1028         new = kmalloc(sizeof(*new), gfp_mask);
1029         if (new == NULL)
1030                 return ERR_PTR(-ENOMEM);
1031         new->sequence = counter;
1032         INIT_LIST_HEAD(&new->list);
1033         new->task = NULL;
1034         return new;
1035 }
1036
1037 void nfs_release_seqid(struct nfs_seqid *seqid)
1038 {
1039         struct nfs_seqid_counter *sequence;
1040
1041         if (seqid == NULL || list_empty(&seqid->list))
1042                 return;
1043         sequence = seqid->sequence;
1044         spin_lock(&sequence->lock);
1045         list_del_init(&seqid->list);
1046         if (!list_empty(&sequence->list)) {
1047                 struct nfs_seqid *next;
1048
1049                 next = list_first_entry(&sequence->list,
1050                                 struct nfs_seqid, list);
1051                 rpc_wake_up_queued_task(&sequence->wait, next->task);
1052         }
1053         spin_unlock(&sequence->lock);
1054 }
1055
1056 void nfs_free_seqid(struct nfs_seqid *seqid)
1057 {
1058         nfs_release_seqid(seqid);
1059         kfree(seqid);
1060 }
1061
1062 /*
1063  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1064  * failed with a seqid incrementing error -
1065  * see comments nfs4.h:seqid_mutating_error()
1066  */
1067 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1068 {
1069         switch (status) {
1070                 case 0:
1071                         break;
1072                 case -NFS4ERR_BAD_SEQID:
1073                         if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1074                                 return;
1075                         pr_warn_ratelimited("NFS: v4 server returned a bad"
1076                                         " sequence-id error on an"
1077                                         " unconfirmed sequence %p!\n",
1078                                         seqid->sequence);
1079                 case -NFS4ERR_STALE_CLIENTID:
1080                 case -NFS4ERR_STALE_STATEID:
1081                 case -NFS4ERR_BAD_STATEID:
1082                 case -NFS4ERR_BADXDR:
1083                 case -NFS4ERR_RESOURCE:
1084                 case -NFS4ERR_NOFILEHANDLE:
1085                         /* Non-seqid mutating errors */
1086                         return;
1087         };
1088         /*
1089          * Note: no locking needed as we are guaranteed to be first
1090          * on the sequence list
1091          */
1092         seqid->sequence->counter++;
1093 }
1094
1095 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1096 {
1097         struct nfs4_state_owner *sp;
1098
1099         if (seqid == NULL)
1100                 return;
1101
1102         sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1103         if (status == -NFS4ERR_BAD_SEQID)
1104                 nfs4_drop_state_owner(sp);
1105         if (!nfs4_has_session(sp->so_server->nfs_client))
1106                 nfs_increment_seqid(status, seqid);
1107 }
1108
1109 /*
1110  * Increment the seqid if the LOCK/LOCKU succeeded, or
1111  * failed with a seqid incrementing error -
1112  * see comments nfs4.h:seqid_mutating_error()
1113  */
1114 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1115 {
1116         if (seqid != NULL)
1117                 nfs_increment_seqid(status, seqid);
1118 }
1119
1120 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1121 {
1122         struct nfs_seqid_counter *sequence;
1123         int status = 0;
1124
1125         if (seqid == NULL)
1126                 goto out;
1127         sequence = seqid->sequence;
1128         spin_lock(&sequence->lock);
1129         seqid->task = task;
1130         if (list_empty(&seqid->list))
1131                 list_add_tail(&seqid->list, &sequence->list);
1132         if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1133                 goto unlock;
1134         rpc_sleep_on(&sequence->wait, task, NULL);
1135         status = -EAGAIN;
1136 unlock:
1137         spin_unlock(&sequence->lock);
1138 out:
1139         return status;
1140 }
1141
1142 static int nfs4_run_state_manager(void *);
1143
1144 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1145 {
1146         smp_mb__before_atomic();
1147         clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1148         smp_mb__after_atomic();
1149         wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1150         rpc_wake_up(&clp->cl_rpcwaitq);
1151 }
1152
1153 /*
1154  * Schedule the nfs_client asynchronous state management routine
1155  */
1156 void nfs4_schedule_state_manager(struct nfs_client *clp)
1157 {
1158         struct task_struct *task;
1159         char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1160
1161         if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1162                 return;
1163         __module_get(THIS_MODULE);
1164         atomic_inc(&clp->cl_count);
1165
1166         /* The rcu_read_lock() is not strictly necessary, as the state
1167          * manager is the only thread that ever changes the rpc_xprt
1168          * after it's initialized.  At this point, we're single threaded. */
1169         rcu_read_lock();
1170         snprintf(buf, sizeof(buf), "%s-manager",
1171                         rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1172         rcu_read_unlock();
1173         task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1174         if (IS_ERR(task)) {
1175                 printk(KERN_ERR "%s: kthread_run: %ld\n",
1176                         __func__, PTR_ERR(task));
1177                 nfs4_clear_state_manager_bit(clp);
1178                 nfs_put_client(clp);
1179                 module_put(THIS_MODULE);
1180         }
1181 }
1182
1183 /*
1184  * Schedule a lease recovery attempt
1185  */
1186 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1187 {
1188         if (!clp)
1189                 return;
1190         if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1191                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1192         dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1193                         clp->cl_hostname);
1194         nfs4_schedule_state_manager(clp);
1195 }
1196 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1197
1198 /**
1199  * nfs4_schedule_migration_recovery - trigger migration recovery
1200  *
1201  * @server: FSID that is migrating
1202  *
1203  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1204  * value is returned.
1205  */
1206 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1207 {
1208         struct nfs_client *clp = server->nfs_client;
1209
1210         if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1211                 pr_err("NFS: volatile file handles not supported (server %s)\n",
1212                                 clp->cl_hostname);
1213                 return -NFS4ERR_IO;
1214         }
1215
1216         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1217                 return -NFS4ERR_IO;
1218
1219         dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1220                         __func__,
1221                         (unsigned long long)server->fsid.major,
1222                         (unsigned long long)server->fsid.minor,
1223                         clp->cl_hostname);
1224
1225         set_bit(NFS_MIG_IN_TRANSITION,
1226                         &((struct nfs_server *)server)->mig_status);
1227         set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1228
1229         nfs4_schedule_state_manager(clp);
1230         return 0;
1231 }
1232 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1233
1234 /**
1235  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1236  *
1237  * @clp: server to check for moved leases
1238  *
1239  */
1240 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1241 {
1242         dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1243                 __func__, clp->cl_clientid, clp->cl_hostname);
1244
1245         set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1246         nfs4_schedule_state_manager(clp);
1247 }
1248 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1249
1250 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1251 {
1252         int res;
1253
1254         might_sleep();
1255
1256         atomic_inc(&clp->cl_count);
1257         res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1258                                  nfs_wait_bit_killable, TASK_KILLABLE);
1259         if (res)
1260                 goto out;
1261         if (clp->cl_cons_state < 0)
1262                 res = clp->cl_cons_state;
1263 out:
1264         nfs_put_client(clp);
1265         return res;
1266 }
1267
1268 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1269 {
1270         unsigned int loop;
1271         int ret;
1272
1273         for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1274                 ret = nfs4_wait_clnt_recover(clp);
1275                 if (ret != 0)
1276                         break;
1277                 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1278                     !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1279                         break;
1280                 nfs4_schedule_state_manager(clp);
1281                 ret = -EIO;
1282         }
1283         return ret;
1284 }
1285
1286 /*
1287  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1288  * @clp: client to process
1289  *
1290  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1291  * resend of the SETCLIENTID and hence re-establish the
1292  * callback channel. Then return all existing delegations.
1293  */
1294 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1295 {
1296         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1297         nfs_expire_all_delegations(clp);
1298         dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1299                         clp->cl_hostname);
1300 }
1301
1302 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1303 {
1304         nfs40_handle_cb_pathdown(clp);
1305         nfs4_schedule_state_manager(clp);
1306 }
1307
1308 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1309 {
1310
1311         if (!nfs4_valid_open_stateid(state))
1312                 return 0;
1313         set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1314         /* Don't recover state that expired before the reboot */
1315         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1316                 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1317                 return 0;
1318         }
1319         set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1320         set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1321         return 1;
1322 }
1323
1324 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1325 {
1326         if (!nfs4_valid_open_stateid(state))
1327                 return 0;
1328         set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1329         clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1330         set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1331         set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1332         return 1;
1333 }
1334
1335 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1336 {
1337         struct nfs_client *clp = server->nfs_client;
1338
1339         if (!nfs4_state_mark_reclaim_nograce(clp, state))
1340                 return -EBADF;
1341         dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1342                         clp->cl_hostname);
1343         nfs4_schedule_state_manager(clp);
1344         return 0;
1345 }
1346 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1347
1348 static struct nfs4_lock_state *
1349 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1350                 const nfs4_stateid *stateid)
1351 {
1352         struct nfs4_lock_state *pos;
1353
1354         list_for_each_entry(pos, &state->lock_states, ls_locks) {
1355                 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1356                         continue;
1357                 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1358                         return pos;
1359         }
1360         return NULL;
1361 }
1362
1363 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1364                 const nfs4_stateid *stateid)
1365 {
1366         bool found = false;
1367
1368         if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1369                 spin_lock(&state->state_lock);
1370                 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1371                         found = true;
1372                 spin_unlock(&state->state_lock);
1373         }
1374         return found;
1375 }
1376
1377 void nfs_inode_find_state_and_recover(struct inode *inode,
1378                 const nfs4_stateid *stateid)
1379 {
1380         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1381         struct nfs_inode *nfsi = NFS_I(inode);
1382         struct nfs_open_context *ctx;
1383         struct nfs4_state *state;
1384         bool found = false;
1385
1386         spin_lock(&inode->i_lock);
1387         list_for_each_entry(ctx, &nfsi->open_files, list) {
1388                 state = ctx->state;
1389                 if (state == NULL)
1390                         continue;
1391                 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1392                     nfs4_state_mark_reclaim_nograce(clp, state)) {
1393                         found = true;
1394                         continue;
1395                 }
1396                 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1397                     nfs4_state_mark_reclaim_nograce(clp, state))
1398                         found = true;
1399         }
1400         spin_unlock(&inode->i_lock);
1401
1402         nfs_inode_find_delegation_state_and_recover(inode, stateid);
1403         if (found)
1404                 nfs4_schedule_state_manager(clp);
1405 }
1406
1407 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1408 {
1409         struct inode *inode = state->inode;
1410         struct nfs_inode *nfsi = NFS_I(inode);
1411         struct nfs_open_context *ctx;
1412
1413         spin_lock(&inode->i_lock);
1414         list_for_each_entry(ctx, &nfsi->open_files, list) {
1415                 if (ctx->state != state)
1416                         continue;
1417                 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1418         }
1419         spin_unlock(&inode->i_lock);
1420 }
1421
1422 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1423 {
1424         set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1425         nfs4_state_mark_open_context_bad(state);
1426 }
1427
1428
1429 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1430 {
1431         struct inode *inode = state->inode;
1432         struct nfs_inode *nfsi = NFS_I(inode);
1433         struct file_lock *fl;
1434         int status = 0;
1435         struct file_lock_context *flctx = inode->i_flctx;
1436         struct list_head *list;
1437
1438         if (flctx == NULL)
1439                 return 0;
1440
1441         list = &flctx->flc_posix;
1442
1443         /* Guard against delegation returns and new lock/unlock calls */
1444         down_write(&nfsi->rwsem);
1445         spin_lock(&flctx->flc_lock);
1446 restart:
1447         list_for_each_entry(fl, list, fl_list) {
1448                 if (nfs_file_open_context(fl->fl_file)->state != state)
1449                         continue;
1450                 spin_unlock(&flctx->flc_lock);
1451                 status = ops->recover_lock(state, fl);
1452                 switch (status) {
1453                 case 0:
1454                         break;
1455                 case -ESTALE:
1456                 case -NFS4ERR_ADMIN_REVOKED:
1457                 case -NFS4ERR_STALE_STATEID:
1458                 case -NFS4ERR_BAD_STATEID:
1459                 case -NFS4ERR_EXPIRED:
1460                 case -NFS4ERR_NO_GRACE:
1461                 case -NFS4ERR_STALE_CLIENTID:
1462                 case -NFS4ERR_BADSESSION:
1463                 case -NFS4ERR_BADSLOT:
1464                 case -NFS4ERR_BAD_HIGH_SLOT:
1465                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1466                         goto out;
1467                 default:
1468                         pr_err("NFS: %s: unhandled error %d\n",
1469                                         __func__, status);
1470                 case -ENOMEM:
1471                 case -NFS4ERR_DENIED:
1472                 case -NFS4ERR_RECLAIM_BAD:
1473                 case -NFS4ERR_RECLAIM_CONFLICT:
1474                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1475                         status = 0;
1476                 }
1477                 spin_lock(&flctx->flc_lock);
1478         }
1479         if (list == &flctx->flc_posix) {
1480                 list = &flctx->flc_flock;
1481                 goto restart;
1482         }
1483         spin_unlock(&flctx->flc_lock);
1484 out:
1485         up_write(&nfsi->rwsem);
1486         return status;
1487 }
1488
1489 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1490 {
1491         struct nfs4_state *state;
1492         struct nfs4_lock_state *lock;
1493         int status = 0;
1494
1495         /* Note: we rely on the sp->so_states list being ordered
1496          * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1497          * states first.
1498          * This is needed to ensure that the server won't give us any
1499          * read delegations that we have to return if, say, we are
1500          * recovering after a network partition or a reboot from a
1501          * server that doesn't support a grace period.
1502          */
1503         spin_lock(&sp->so_lock);
1504         raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1505 restart:
1506         list_for_each_entry(state, &sp->so_states, open_states) {
1507                 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1508                         continue;
1509                 if (!nfs4_valid_open_stateid(state))
1510                         continue;
1511                 if (state->state == 0)
1512                         continue;
1513                 atomic_inc(&state->count);
1514                 spin_unlock(&sp->so_lock);
1515                 status = ops->recover_open(sp, state);
1516                 if (status >= 0) {
1517                         status = nfs4_reclaim_locks(state, ops);
1518                         if (status >= 0) {
1519                                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1520                                         spin_lock(&state->state_lock);
1521                                         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1522                                                 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1523                                                         pr_warn_ratelimited("NFS: "
1524                                                                             "%s: Lock reclaim "
1525                                                                             "failed!\n", __func__);
1526                                         }
1527                                         spin_unlock(&state->state_lock);
1528                                 }
1529                                 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1530                                         &state->flags);
1531                                 nfs4_put_open_state(state);
1532                                 spin_lock(&sp->so_lock);
1533                                 goto restart;
1534                         }
1535                 }
1536                 switch (status) {
1537                         default:
1538                                 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1539                                         __func__, status);
1540                         case -ENOENT:
1541                         case -ENOMEM:
1542                         case -EACCES:
1543                         case -EROFS:
1544                         case -EIO:
1545                         case -ESTALE:
1546                                 /* Open state on this file cannot be recovered */
1547                                 nfs4_state_mark_recovery_failed(state, status);
1548                                 break;
1549                         case -EAGAIN:
1550                                 ssleep(1);
1551                         case -NFS4ERR_ADMIN_REVOKED:
1552                         case -NFS4ERR_STALE_STATEID:
1553                         case -NFS4ERR_OLD_STATEID:
1554                         case -NFS4ERR_BAD_STATEID:
1555                         case -NFS4ERR_RECLAIM_BAD:
1556                         case -NFS4ERR_RECLAIM_CONFLICT:
1557                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1558                                 break;
1559                         case -NFS4ERR_EXPIRED:
1560                         case -NFS4ERR_NO_GRACE:
1561                                 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1562                         case -NFS4ERR_STALE_CLIENTID:
1563                         case -NFS4ERR_BADSESSION:
1564                         case -NFS4ERR_BADSLOT:
1565                         case -NFS4ERR_BAD_HIGH_SLOT:
1566                         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1567                                 goto out_err;
1568                 }
1569                 nfs4_put_open_state(state);
1570                 spin_lock(&sp->so_lock);
1571                 goto restart;
1572         }
1573         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1574         spin_unlock(&sp->so_lock);
1575         return 0;
1576 out_err:
1577         nfs4_put_open_state(state);
1578         spin_lock(&sp->so_lock);
1579         raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1580         spin_unlock(&sp->so_lock);
1581         return status;
1582 }
1583
1584 static void nfs4_clear_open_state(struct nfs4_state *state)
1585 {
1586         struct nfs4_lock_state *lock;
1587
1588         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1589         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1590         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1591         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1592         spin_lock(&state->state_lock);
1593         list_for_each_entry(lock, &state->lock_states, ls_locks) {
1594                 lock->ls_seqid.flags = 0;
1595                 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1596         }
1597         spin_unlock(&state->state_lock);
1598 }
1599
1600 static void nfs4_reset_seqids(struct nfs_server *server,
1601         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1602 {
1603         struct nfs_client *clp = server->nfs_client;
1604         struct nfs4_state_owner *sp;
1605         struct rb_node *pos;
1606         struct nfs4_state *state;
1607
1608         spin_lock(&clp->cl_lock);
1609         for (pos = rb_first(&server->state_owners);
1610              pos != NULL;
1611              pos = rb_next(pos)) {
1612                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1613                 sp->so_seqid.flags = 0;
1614                 spin_lock(&sp->so_lock);
1615                 list_for_each_entry(state, &sp->so_states, open_states) {
1616                         if (mark_reclaim(clp, state))
1617                                 nfs4_clear_open_state(state);
1618                 }
1619                 spin_unlock(&sp->so_lock);
1620         }
1621         spin_unlock(&clp->cl_lock);
1622 }
1623
1624 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1625         int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1626 {
1627         struct nfs_server *server;
1628
1629         rcu_read_lock();
1630         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1631                 nfs4_reset_seqids(server, mark_reclaim);
1632         rcu_read_unlock();
1633 }
1634
1635 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1636 {
1637         /* Mark all delegations for reclaim */
1638         nfs_delegation_mark_reclaim(clp);
1639         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1640 }
1641
1642 static void nfs4_reclaim_complete(struct nfs_client *clp,
1643                                  const struct nfs4_state_recovery_ops *ops,
1644                                  struct rpc_cred *cred)
1645 {
1646         /* Notify the server we're done reclaiming our state */
1647         if (ops->reclaim_complete)
1648                 (void)ops->reclaim_complete(clp, cred);
1649 }
1650
1651 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1652 {
1653         struct nfs_client *clp = server->nfs_client;
1654         struct nfs4_state_owner *sp;
1655         struct rb_node *pos;
1656         struct nfs4_state *state;
1657
1658         spin_lock(&clp->cl_lock);
1659         for (pos = rb_first(&server->state_owners);
1660              pos != NULL;
1661              pos = rb_next(pos)) {
1662                 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1663                 spin_lock(&sp->so_lock);
1664                 list_for_each_entry(state, &sp->so_states, open_states) {
1665                         if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1666                                                 &state->flags))
1667                                 continue;
1668                         nfs4_state_mark_reclaim_nograce(clp, state);
1669                 }
1670                 spin_unlock(&sp->so_lock);
1671         }
1672         spin_unlock(&clp->cl_lock);
1673 }
1674
1675 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1676 {
1677         struct nfs_server *server;
1678
1679         if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1680                 return 0;
1681
1682         rcu_read_lock();
1683         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1684                 nfs4_clear_reclaim_server(server);
1685         rcu_read_unlock();
1686
1687         nfs_delegation_reap_unclaimed(clp);
1688         return 1;
1689 }
1690
1691 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1692 {
1693         const struct nfs4_state_recovery_ops *ops;
1694         struct rpc_cred *cred;
1695
1696         if (!nfs4_state_clear_reclaim_reboot(clp))
1697                 return;
1698         ops = clp->cl_mvops->reboot_recovery_ops;
1699         cred = nfs4_get_clid_cred(clp);
1700         nfs4_reclaim_complete(clp, ops, cred);
1701         put_rpccred(cred);
1702 }
1703
1704 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1705 {
1706         nfs_mark_test_expired_all_delegations(clp);
1707         nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1708 }
1709
1710 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1711 {
1712         switch (error) {
1713                 case 0:
1714                         break;
1715                 case -NFS4ERR_CB_PATH_DOWN:
1716                         nfs40_handle_cb_pathdown(clp);
1717                         break;
1718                 case -NFS4ERR_NO_GRACE:
1719                         nfs4_state_end_reclaim_reboot(clp);
1720                         break;
1721                 case -NFS4ERR_STALE_CLIENTID:
1722                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1723                         nfs4_state_clear_reclaim_reboot(clp);
1724                         nfs4_state_start_reclaim_reboot(clp);
1725                         break;
1726                 case -NFS4ERR_EXPIRED:
1727                         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1728                         nfs4_state_start_reclaim_nograce(clp);
1729                         break;
1730                 case -NFS4ERR_BADSESSION:
1731                 case -NFS4ERR_BADSLOT:
1732                 case -NFS4ERR_BAD_HIGH_SLOT:
1733                 case -NFS4ERR_DEADSESSION:
1734                 case -NFS4ERR_SEQ_FALSE_RETRY:
1735                 case -NFS4ERR_SEQ_MISORDERED:
1736                         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1737                         /* Zero session reset errors */
1738                         break;
1739                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1740                         set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1741                         break;
1742                 default:
1743                         dprintk("%s: failed to handle error %d for server %s\n",
1744                                         __func__, error, clp->cl_hostname);
1745                         return error;
1746         }
1747         dprintk("%s: handled error %d for server %s\n", __func__, error,
1748                         clp->cl_hostname);
1749         return 0;
1750 }
1751
1752 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1753 {
1754         struct nfs4_state_owner *sp;
1755         struct nfs_server *server;
1756         struct rb_node *pos;
1757         int status = 0;
1758
1759 restart:
1760         rcu_read_lock();
1761         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1762                 nfs4_purge_state_owners(server);
1763                 spin_lock(&clp->cl_lock);
1764                 for (pos = rb_first(&server->state_owners);
1765                      pos != NULL;
1766                      pos = rb_next(pos)) {
1767                         sp = rb_entry(pos,
1768                                 struct nfs4_state_owner, so_server_node);
1769                         if (!test_and_clear_bit(ops->owner_flag_bit,
1770                                                         &sp->so_flags))
1771                                 continue;
1772                         if (!atomic_inc_not_zero(&sp->so_count))
1773                                 continue;
1774                         spin_unlock(&clp->cl_lock);
1775                         rcu_read_unlock();
1776
1777                         status = nfs4_reclaim_open_state(sp, ops);
1778                         if (status < 0) {
1779                                 set_bit(ops->owner_flag_bit, &sp->so_flags);
1780                                 nfs4_put_state_owner(sp);
1781                                 status = nfs4_recovery_handle_error(clp, status);
1782                                 return (status != 0) ? status : -EAGAIN;
1783                         }
1784
1785                         nfs4_put_state_owner(sp);
1786                         goto restart;
1787                 }
1788                 spin_unlock(&clp->cl_lock);
1789         }
1790         rcu_read_unlock();
1791         return 0;
1792 }
1793
1794 static int nfs4_check_lease(struct nfs_client *clp)
1795 {
1796         struct rpc_cred *cred;
1797         const struct nfs4_state_maintenance_ops *ops =
1798                 clp->cl_mvops->state_renewal_ops;
1799         int status;
1800
1801         /* Is the client already known to have an expired lease? */
1802         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1803                 return 0;
1804         spin_lock(&clp->cl_lock);
1805         cred = ops->get_state_renewal_cred_locked(clp);
1806         spin_unlock(&clp->cl_lock);
1807         if (cred == NULL) {
1808                 cred = nfs4_get_clid_cred(clp);
1809                 status = -ENOKEY;
1810                 if (cred == NULL)
1811                         goto out;
1812         }
1813         status = ops->renew_lease(clp, cred);
1814         put_rpccred(cred);
1815         if (status == -ETIMEDOUT) {
1816                 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1817                 return 0;
1818         }
1819 out:
1820         return nfs4_recovery_handle_error(clp, status);
1821 }
1822
1823 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1824  * and for recoverable errors on EXCHANGE_ID for v4.1
1825  */
1826 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1827 {
1828         switch (status) {
1829         case -NFS4ERR_SEQ_MISORDERED:
1830                 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1831                         return -ESERVERFAULT;
1832                 /* Lease confirmation error: retry after purging the lease */
1833                 ssleep(1);
1834                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1835                 break;
1836         case -NFS4ERR_STALE_CLIENTID:
1837                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1838                 nfs4_state_start_reclaim_reboot(clp);
1839                 break;
1840         case -NFS4ERR_CLID_INUSE:
1841                 pr_err("NFS: Server %s reports our clientid is in use\n",
1842                         clp->cl_hostname);
1843                 nfs_mark_client_ready(clp, -EPERM);
1844                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1845                 return -EPERM;
1846         case -EACCES:
1847         case -NFS4ERR_DELAY:
1848         case -ETIMEDOUT:
1849         case -EAGAIN:
1850                 ssleep(1);
1851                 break;
1852
1853         case -NFS4ERR_MINOR_VERS_MISMATCH:
1854                 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1855                         nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1856                 dprintk("%s: exit with error %d for server %s\n",
1857                                 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1858                 return -EPROTONOSUPPORT;
1859         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1860                                  * in nfs4_exchange_id */
1861         default:
1862                 dprintk("%s: exit with error %d for server %s\n", __func__,
1863                                 status, clp->cl_hostname);
1864                 return status;
1865         }
1866         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1867         dprintk("%s: handled error %d for server %s\n", __func__, status,
1868                         clp->cl_hostname);
1869         return 0;
1870 }
1871
1872 static int nfs4_establish_lease(struct nfs_client *clp)
1873 {
1874         struct rpc_cred *cred;
1875         const struct nfs4_state_recovery_ops *ops =
1876                 clp->cl_mvops->reboot_recovery_ops;
1877         int status;
1878
1879         nfs4_begin_drain_session(clp);
1880         cred = nfs4_get_clid_cred(clp);
1881         if (cred == NULL)
1882                 return -ENOENT;
1883         status = ops->establish_clid(clp, cred);
1884         put_rpccred(cred);
1885         if (status != 0)
1886                 return status;
1887         pnfs_destroy_all_layouts(clp);
1888         return 0;
1889 }
1890
1891 /*
1892  * Returns zero or a negative errno.  NFS4ERR values are converted
1893  * to local errno values.
1894  */
1895 static int nfs4_reclaim_lease(struct nfs_client *clp)
1896 {
1897         int status;
1898
1899         status = nfs4_establish_lease(clp);
1900         if (status < 0)
1901                 return nfs4_handle_reclaim_lease_error(clp, status);
1902         if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1903                 nfs4_state_start_reclaim_nograce(clp);
1904         if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1905                 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1906         clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1907         clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1908         return 0;
1909 }
1910
1911 static int nfs4_purge_lease(struct nfs_client *clp)
1912 {
1913         int status;
1914
1915         status = nfs4_establish_lease(clp);
1916         if (status < 0)
1917                 return nfs4_handle_reclaim_lease_error(clp, status);
1918         clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1919         set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1920         nfs4_state_start_reclaim_nograce(clp);
1921         return 0;
1922 }
1923
1924 /*
1925  * Try remote migration of one FSID from a source server to a
1926  * destination server.  The source server provides a list of
1927  * potential destinations.
1928  *
1929  * Returns zero or a negative NFS4ERR status code.
1930  */
1931 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1932 {
1933         struct nfs_client *clp = server->nfs_client;
1934         struct nfs4_fs_locations *locations = NULL;
1935         struct inode *inode;
1936         struct page *page;
1937         int status, result;
1938
1939         dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1940                         (unsigned long long)server->fsid.major,
1941                         (unsigned long long)server->fsid.minor,
1942                         clp->cl_hostname);
1943
1944         result = 0;
1945         page = alloc_page(GFP_KERNEL);
1946         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1947         if (page == NULL || locations == NULL) {
1948                 dprintk("<-- %s: no memory\n", __func__);
1949                 goto out;
1950         }
1951
1952         inode = d_inode(server->super->s_root);
1953         result = nfs4_proc_get_locations(inode, locations, page, cred);
1954         if (result) {
1955                 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1956                         __func__, result);
1957                 goto out;
1958         }
1959
1960         result = -NFS4ERR_NXIO;
1961         if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1962                 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1963                         __func__);
1964                 goto out;
1965         }
1966
1967         nfs4_begin_drain_session(clp);
1968
1969         status = nfs4_replace_transport(server, locations);
1970         if (status != 0) {
1971                 dprintk("<-- %s: failed to replace transport: %d\n",
1972                         __func__, status);
1973                 goto out;
1974         }
1975
1976         result = 0;
1977         dprintk("<-- %s: migration succeeded\n", __func__);
1978
1979 out:
1980         if (page != NULL)
1981                 __free_page(page);
1982         kfree(locations);
1983         if (result) {
1984                 pr_err("NFS: migration recovery failed (server %s)\n",
1985                                 clp->cl_hostname);
1986                 set_bit(NFS_MIG_FAILED, &server->mig_status);
1987         }
1988         return result;
1989 }
1990
1991 /*
1992  * Returns zero or a negative NFS4ERR status code.
1993  */
1994 static int nfs4_handle_migration(struct nfs_client *clp)
1995 {
1996         const struct nfs4_state_maintenance_ops *ops =
1997                                 clp->cl_mvops->state_renewal_ops;
1998         struct nfs_server *server;
1999         struct rpc_cred *cred;
2000
2001         dprintk("%s: migration reported on \"%s\"\n", __func__,
2002                         clp->cl_hostname);
2003
2004         spin_lock(&clp->cl_lock);
2005         cred = ops->get_state_renewal_cred_locked(clp);
2006         spin_unlock(&clp->cl_lock);
2007         if (cred == NULL)
2008                 return -NFS4ERR_NOENT;
2009
2010         clp->cl_mig_gen++;
2011 restart:
2012         rcu_read_lock();
2013         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2014                 int status;
2015
2016                 if (server->mig_gen == clp->cl_mig_gen)
2017                         continue;
2018                 server->mig_gen = clp->cl_mig_gen;
2019
2020                 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2021                                                 &server->mig_status))
2022                         continue;
2023
2024                 rcu_read_unlock();
2025                 status = nfs4_try_migration(server, cred);
2026                 if (status < 0) {
2027                         put_rpccred(cred);
2028                         return status;
2029                 }
2030                 goto restart;
2031         }
2032         rcu_read_unlock();
2033         put_rpccred(cred);
2034         return 0;
2035 }
2036
2037 /*
2038  * Test each nfs_server on the clp's cl_superblocks list to see
2039  * if it's moved to another server.  Stop when the server no longer
2040  * returns NFS4ERR_LEASE_MOVED.
2041  */
2042 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2043 {
2044         const struct nfs4_state_maintenance_ops *ops =
2045                                 clp->cl_mvops->state_renewal_ops;
2046         struct nfs_server *server;
2047         struct rpc_cred *cred;
2048
2049         dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2050                         clp->cl_hostname);
2051
2052         spin_lock(&clp->cl_lock);
2053         cred = ops->get_state_renewal_cred_locked(clp);
2054         spin_unlock(&clp->cl_lock);
2055         if (cred == NULL)
2056                 return -NFS4ERR_NOENT;
2057
2058         clp->cl_mig_gen++;
2059 restart:
2060         rcu_read_lock();
2061         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2062                 struct inode *inode;
2063                 int status;
2064
2065                 if (server->mig_gen == clp->cl_mig_gen)
2066                         continue;
2067                 server->mig_gen = clp->cl_mig_gen;
2068
2069                 rcu_read_unlock();
2070
2071                 inode = d_inode(server->super->s_root);
2072                 status = nfs4_proc_fsid_present(inode, cred);
2073                 if (status != -NFS4ERR_MOVED)
2074                         goto restart;   /* wasn't this one */
2075                 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2076                         goto restart;   /* there are more */
2077                 goto out;
2078         }
2079         rcu_read_unlock();
2080
2081 out:
2082         put_rpccred(cred);
2083         return 0;
2084 }
2085
2086 /**
2087  * nfs4_discover_server_trunking - Detect server IP address trunking
2088  *
2089  * @clp: nfs_client under test
2090  * @result: OUT: found nfs_client, or clp
2091  *
2092  * Returns zero or a negative errno.  If zero is returned,
2093  * an nfs_client pointer is planted in "result".
2094  *
2095  * Note: since we are invoked in process context, and
2096  * not from inside the state manager, we cannot use
2097  * nfs4_handle_reclaim_lease_error().
2098  */
2099 int nfs4_discover_server_trunking(struct nfs_client *clp,
2100                                   struct nfs_client **result)
2101 {
2102         const struct nfs4_state_recovery_ops *ops =
2103                                 clp->cl_mvops->reboot_recovery_ops;
2104         struct rpc_clnt *clnt;
2105         struct rpc_cred *cred;
2106         int i, status;
2107
2108         dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2109
2110         clnt = clp->cl_rpcclient;
2111         i = 0;
2112
2113         mutex_lock(&nfs_clid_init_mutex);
2114 again:
2115         status  = -ENOENT;
2116         cred = nfs4_get_clid_cred(clp);
2117         if (cred == NULL)
2118                 goto out_unlock;
2119
2120         status = ops->detect_trunking(clp, result, cred);
2121         put_rpccred(cred);
2122         switch (status) {
2123         case 0:
2124                 break;
2125         case -ETIMEDOUT:
2126                 if (clnt->cl_softrtry)
2127                         break;
2128         case -NFS4ERR_DELAY:
2129         case -EAGAIN:
2130                 ssleep(1);
2131         case -NFS4ERR_STALE_CLIENTID:
2132                 dprintk("NFS: %s after status %d, retrying\n",
2133                         __func__, status);
2134                 goto again;
2135         case -EACCES:
2136                 if (i++ == 0) {
2137                         nfs4_root_machine_cred(clp);
2138                         goto again;
2139                 }
2140                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2141                         break;
2142         case -NFS4ERR_CLID_INUSE:
2143         case -NFS4ERR_WRONGSEC:
2144                 /* No point in retrying if we already used RPC_AUTH_UNIX */
2145                 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2146                         status = -EPERM;
2147                         break;
2148                 }
2149                 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2150                 if (IS_ERR(clnt)) {
2151                         status = PTR_ERR(clnt);
2152                         break;
2153                 }
2154                 /* Note: this is safe because we haven't yet marked the
2155                  * client as ready, so we are the only user of
2156                  * clp->cl_rpcclient
2157                  */
2158                 clnt = xchg(&clp->cl_rpcclient, clnt);
2159                 rpc_shutdown_client(clnt);
2160                 clnt = clp->cl_rpcclient;
2161                 goto again;
2162
2163         case -NFS4ERR_MINOR_VERS_MISMATCH:
2164                 status = -EPROTONOSUPPORT;
2165                 break;
2166
2167         case -EKEYEXPIRED:
2168         case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2169                                  * in nfs4_exchange_id */
2170                 status = -EKEYEXPIRED;
2171                 break;
2172         default:
2173                 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2174                                 __func__, status);
2175                 status = -EIO;
2176         }
2177
2178 out_unlock:
2179         mutex_unlock(&nfs_clid_init_mutex);
2180         dprintk("NFS: %s: status = %d\n", __func__, status);
2181         return status;
2182 }
2183
2184 #ifdef CONFIG_NFS_V4_1
2185 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2186 {
2187         struct nfs_client *clp = session->clp;
2188
2189         switch (err) {
2190         default:
2191                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2192                 break;
2193         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2194                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2195         }
2196         nfs4_schedule_state_manager(clp);
2197 }
2198 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2199
2200 void nfs41_notify_server(struct nfs_client *clp)
2201 {
2202         /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2203         set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2204         nfs4_schedule_state_manager(clp);
2205 }
2206
2207 static void nfs4_reset_all_state(struct nfs_client *clp)
2208 {
2209         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2210                 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2211                 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2212                 nfs4_state_start_reclaim_nograce(clp);
2213                 dprintk("%s: scheduling reset of all state for server %s!\n",
2214                                 __func__, clp->cl_hostname);
2215                 nfs4_schedule_state_manager(clp);
2216         }
2217 }
2218
2219 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2220 {
2221         if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2222                 nfs4_state_start_reclaim_reboot(clp);
2223                 dprintk("%s: server %s rebooted!\n", __func__,
2224                                 clp->cl_hostname);
2225                 nfs4_schedule_state_manager(clp);
2226         }
2227 }
2228
2229 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2230 {
2231         nfs4_reset_all_state(clp);
2232         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2233 }
2234
2235 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2236 {
2237         nfs4_state_start_reclaim_nograce(clp);
2238         nfs4_schedule_state_manager(clp);
2239
2240         dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2241 }
2242
2243 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2244 {
2245         /* FIXME: For now, we destroy all layouts. */
2246         pnfs_destroy_all_layouts(clp);
2247         /* FIXME: For now, we test all delegations+open state+locks. */
2248         nfs41_handle_some_state_revoked(clp);
2249         dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2250                         clp->cl_hostname);
2251 }
2252
2253 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2254 {
2255         set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2256         nfs4_schedule_state_manager(clp);
2257
2258         dprintk("%s: server %s declared a backchannel fault\n", __func__,
2259                         clp->cl_hostname);
2260 }
2261
2262 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2263 {
2264         if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2265                 &clp->cl_state) == 0)
2266                 nfs4_schedule_state_manager(clp);
2267 }
2268
2269 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2270                 bool recovery)
2271 {
2272         if (!flags)
2273                 return;
2274
2275         dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2276                 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2277         /*
2278          * If we're called from the state manager thread, then assume we're
2279          * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2280          * Those flags are expected to remain set until we're done
2281          * recovering (see RFC5661, section 18.46.3).
2282          */
2283         if (recovery)
2284                 goto out_recovery;
2285
2286         if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2287                 nfs41_handle_server_reboot(clp);
2288         if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2289                 nfs41_handle_all_state_revoked(clp);
2290         if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2291                             SEQ4_STATUS_ADMIN_STATE_REVOKED))
2292                 nfs41_handle_some_state_revoked(clp);
2293         if (flags & SEQ4_STATUS_LEASE_MOVED)
2294                 nfs4_schedule_lease_moved_recovery(clp);
2295         if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2296                 nfs41_handle_recallable_state_revoked(clp);
2297 out_recovery:
2298         if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2299                 nfs41_handle_backchannel_fault(clp);
2300         else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2301                                 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2302                 nfs41_handle_cb_path_down(clp);
2303 }
2304
2305 static int nfs4_reset_session(struct nfs_client *clp)
2306 {
2307         struct rpc_cred *cred;
2308         int status;
2309
2310         if (!nfs4_has_session(clp))
2311                 return 0;
2312         nfs4_begin_drain_session(clp);
2313         cred = nfs4_get_clid_cred(clp);
2314         status = nfs4_proc_destroy_session(clp->cl_session, cred);
2315         switch (status) {
2316         case 0:
2317         case -NFS4ERR_BADSESSION:
2318         case -NFS4ERR_DEADSESSION:
2319                 break;
2320         case -NFS4ERR_BACK_CHAN_BUSY:
2321         case -NFS4ERR_DELAY:
2322                 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2323                 status = 0;
2324                 ssleep(1);
2325                 goto out;
2326         default:
2327                 status = nfs4_recovery_handle_error(clp, status);
2328                 goto out;
2329         }
2330
2331         memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2332         status = nfs4_proc_create_session(clp, cred);
2333         if (status) {
2334                 dprintk("%s: session reset failed with status %d for server %s!\n",
2335                         __func__, status, clp->cl_hostname);
2336                 status = nfs4_handle_reclaim_lease_error(clp, status);
2337                 goto out;
2338         }
2339         nfs41_finish_session_reset(clp);
2340         dprintk("%s: session reset was successful for server %s!\n",
2341                         __func__, clp->cl_hostname);
2342 out:
2343         if (cred)
2344                 put_rpccred(cred);
2345         return status;
2346 }
2347
2348 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2349 {
2350         struct rpc_cred *cred;
2351         int ret;
2352
2353         if (!nfs4_has_session(clp))
2354                 return 0;
2355         nfs4_begin_drain_session(clp);
2356         cred = nfs4_get_clid_cred(clp);
2357         ret = nfs4_proc_bind_conn_to_session(clp, cred);
2358         if (cred)
2359                 put_rpccred(cred);
2360         clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2361         switch (ret) {
2362         case 0:
2363                 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2364                         __func__, clp->cl_hostname);
2365                 break;
2366         case -NFS4ERR_DELAY:
2367                 ssleep(1);
2368                 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2369                 break;
2370         default:
2371                 return nfs4_recovery_handle_error(clp, ret);
2372         }
2373         return 0;
2374 }
2375 #else /* CONFIG_NFS_V4_1 */
2376 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2377
2378 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2379 {
2380         return 0;
2381 }
2382 #endif /* CONFIG_NFS_V4_1 */
2383
2384 static void nfs4_state_manager(struct nfs_client *clp)
2385 {
2386         int status = 0;
2387         const char *section = "", *section_sep = "";
2388
2389         /* Ensure exclusive access to NFSv4 state */
2390         do {
2391                 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2392                         section = "purge state";
2393                         status = nfs4_purge_lease(clp);
2394                         if (status < 0)
2395                                 goto out_error;
2396                         continue;
2397                 }
2398
2399                 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2400                         section = "lease expired";
2401                         /* We're going to have to re-establish a clientid */
2402                         status = nfs4_reclaim_lease(clp);
2403                         if (status < 0)
2404                                 goto out_error;
2405                         continue;
2406                 }
2407
2408                 /* Initialize or reset the session */
2409                 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2410                         section = "reset session";
2411                         status = nfs4_reset_session(clp);
2412                         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2413                                 continue;
2414                         if (status < 0)
2415                                 goto out_error;
2416                 }
2417
2418                 /* Send BIND_CONN_TO_SESSION */
2419                 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2420                                 &clp->cl_state)) {
2421                         section = "bind conn to session";
2422                         status = nfs4_bind_conn_to_session(clp);
2423                         if (status < 0)
2424                                 goto out_error;
2425                         continue;
2426                 }
2427
2428                 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2429                         section = "check lease";
2430                         status = nfs4_check_lease(clp);
2431                         if (status < 0)
2432                                 goto out_error;
2433                         continue;
2434                 }
2435
2436                 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2437                         section = "migration";
2438                         status = nfs4_handle_migration(clp);
2439                         if (status < 0)
2440                                 goto out_error;
2441                 }
2442
2443                 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2444                         section = "lease moved";
2445                         status = nfs4_handle_lease_moved(clp);
2446                         if (status < 0)
2447                                 goto out_error;
2448                 }
2449
2450                 /* First recover reboot state... */
2451                 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2452                         section = "reclaim reboot";
2453                         status = nfs4_do_reclaim(clp,
2454                                 clp->cl_mvops->reboot_recovery_ops);
2455                         if (status == -EAGAIN)
2456                                 continue;
2457                         if (status < 0)
2458                                 goto out_error;
2459                         nfs4_state_end_reclaim_reboot(clp);
2460                 }
2461
2462                 /* Detect expired delegations... */
2463                 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2464                         section = "detect expired delegations";
2465                         nfs_reap_expired_delegations(clp);
2466                         continue;
2467                 }
2468
2469                 /* Now recover expired state... */
2470                 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2471                         section = "reclaim nograce";
2472                         status = nfs4_do_reclaim(clp,
2473                                 clp->cl_mvops->nograce_recovery_ops);
2474                         if (status == -EAGAIN)
2475                                 continue;
2476                         if (status < 0)
2477                                 goto out_error;
2478                 }
2479
2480                 nfs4_end_drain_session(clp);
2481                 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2482                         nfs_client_return_marked_delegations(clp);
2483                         continue;
2484                 }
2485
2486                 nfs4_clear_state_manager_bit(clp);
2487                 /* Did we race with an attempt to give us more work? */
2488                 if (clp->cl_state == 0)
2489                         break;
2490                 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2491                         break;
2492         } while (atomic_read(&clp->cl_count) > 1);
2493         return;
2494 out_error:
2495         if (strlen(section))
2496                 section_sep = ": ";
2497         pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2498                         " with error %d\n", section_sep, section,
2499                         clp->cl_hostname, -status);
2500         ssleep(1);
2501         nfs4_end_drain_session(clp);
2502         nfs4_clear_state_manager_bit(clp);
2503 }
2504
2505 static int nfs4_run_state_manager(void *ptr)
2506 {
2507         struct nfs_client *clp = ptr;
2508
2509         allow_signal(SIGKILL);
2510         nfs4_state_manager(clp);
2511         nfs_put_client(clp);
2512         module_put_and_exit(0);
2513         return 0;
2514 }
2515
2516 /*
2517  * Local variables:
2518  *  c-basic-offset: 8
2519  * End:
2520  */